Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generative Diffusion Contrastive Network for Multi-View Clustering

Jian Zhu, Xin Zou, Xi Wang, Lei Liu, Chang Tang, Li-Rong Dai

TL;DR

This work tackles the problem of low-quality data in multi-view clustering by introducing Stochastic Generative Diffusion Fusion (SGDF) and the Generative Diffusion Contrastive Network (GDCN). SGDF robustly fuses multi-view embeddings through a diffusion-based, multi-sample mechanism that averages multiple generative reconstructions, while GDCN combines per-view autoencoders, SGDF fusion, and contrastive learning to produce a robust common representation for clustering. The model achieves state-of-the-art results on four public MVC datasets, significantly outperforming existing deep MVC methods and demonstrating strong robustness to noisy or missing views. The approach offers a scalable, effective framework for deep MVC and provides public code to facilitate further research and application in multi-view fusion tasks.

Abstract

In recent years, Multi-View Clustering (MVC) has been significantly advanced under the influence of deep learning. By integrating heterogeneous data from multiple views, MVC enhances clustering analysis, making multi-view fusion critical to clustering performance. However, there is a problem of low-quality data in multi-view fusion. This problem primarily arises from two reasons: 1) Certain views are contaminated by noisy data. 2) Some views suffer from missing data. This paper proposes a novel Stochastic Generative Diffusion Fusion (SGDF) method to address this problem. SGDF leverages a multiple generative mechanism for the multi-view feature of each sample. It is robust to low-quality data. Building on SGDF, we further present the Generative Diffusion Contrastive Network (GDCN). Extensive experiments show that GDCN achieves the state-of-the-art results in deep MVC tasks. The source code is publicly available at https://github.com/HackerHyper/GDCN.

Generative Diffusion Contrastive Network for Multi-View Clustering

TL;DR

This work tackles the problem of low-quality data in multi-view clustering by introducing Stochastic Generative Diffusion Fusion (SGDF) and the Generative Diffusion Contrastive Network (GDCN). SGDF robustly fuses multi-view embeddings through a diffusion-based, multi-sample mechanism that averages multiple generative reconstructions, while GDCN combines per-view autoencoders, SGDF fusion, and contrastive learning to produce a robust common representation for clustering. The model achieves state-of-the-art results on four public MVC datasets, significantly outperforming existing deep MVC methods and demonstrating strong robustness to noisy or missing views. The approach offers a scalable, effective framework for deep MVC and provides public code to facilitate further research and application in multi-view fusion tasks.

Abstract

In recent years, Multi-View Clustering (MVC) has been significantly advanced under the influence of deep learning. By integrating heterogeneous data from multiple views, MVC enhances clustering analysis, making multi-view fusion critical to clustering performance. However, there is a problem of low-quality data in multi-view fusion. This problem primarily arises from two reasons: 1) Certain views are contaminated by noisy data. 2) Some views suffer from missing data. This paper proposes a novel Stochastic Generative Diffusion Fusion (SGDF) method to address this problem. SGDF leverages a multiple generative mechanism for the multi-view feature of each sample. It is robust to low-quality data. Building on SGDF, we further present the Generative Diffusion Contrastive Network (GDCN). Extensive experiments show that GDCN achieves the state-of-the-art results in deep MVC tasks. The source code is publicly available at https://github.com/HackerHyper/GDCN.

Paper Structure

This paper contains 13 sections, 13 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. The Proposed Methodology
  3. Autoencoder Module
  4. Stochastic Generative Diffusion Fusion
  5. Diffusion Reverse Process
  6. Multiple Generative Mechanism
  7. Loss Function and Clustering module
  8. Experiments
  9. Experimental Settings
  10. Experimental comparative results
  11. Ablation Study
  12. Convergence, Visualization, and Parameter Analysis.
  13. Conclusion and Future Work

Figures (3)

  • Figure 1: The framework of GDCN. The framework consists of the Autoencoder module, the SGDF module, and the Contrastive Learning (CL) module. The Autoencoder module aims to obtain view-specific representations that effectively reconstruct the original data. The SGDF module performs robust fusion of the multi-view data. Eventually, the Contrastive Learning module is used to get a common representation. K-Means is a module used for clustering.
  • Figure 2: The convergence analysis and visualization analysis on NGs. (a) The training loss/Acc curves; (b) The t-SNE plot of the common representations $\{\hat{h}_i\}_{i=1}^{N}$.
  • Figure 3: The hyper-parameter analysis on NGs.